1
$\begingroup$

I am post doctoral fellow in neurosciences. I have a quite basic question that I failed to solve by myself.

In a pretest experiment for an fMRI study, we have asked to 3 age groups (~30 participants in each group) to listen 6 different musics. Each of the 6 musics is supposed to elicit only one particular emotion. After listening each music, we collected a binary outcome.

To sum up, each participant has 6 binary responses, one for each modality of the music factor.

Intuitively, because the DV is binary and participants answered several "trials", I ran a GEE model. Since I found a significant Music*Age interaction, I ran post-hoc tests sorted by the music factor (via the emmeans package in R) to compare all age groups at each music modality.

After running these analyses, I felt that something was wrong but I am still not clearly able to explain why. Is it correct or are some analyses more appropriate with this design?

Let me know if something should be clarified and thanks for your help!

Improve this question
$\endgroup$
3
  • 1
    $\begingroup$ Hi, I'm no expert of GEE but I can see few alternative ways to go about analyzing your data (for example multivariate logit); which one is the most appropriate depends also on the question you are asking with this experiment. What is your research question here? $\endgroup$
    – matteo
    Commented Sep 6, 2019 at 17:16
  • $\begingroup$ Hi, thanks for answering so promptly. This is exactly what I felt, if I am only interested in comparing age groups at each music modality I can run multivariate logit but if I want also to obtain results on within-groups comparisons I have to run a GEE model? $\endgroup$
    – Hub Fournier
    Commented Sep 6, 2019 at 17:35
  • 1
    $\begingroup$ If you are not interested in the correlation between the responses at the different modalities, but only in comparing age groups, then I think you could also do a mixed-effect logistic regression (e.g. package lme4), with Music and Age as fixed-effects predictors (and random intercept for subjects) $\endgroup$
    – matteo
    Commented Sep 6, 2019 at 18:19

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.